News Hub

Report: Microsoft handed OpenAI $500m in Azure credits

Written by Tue 18 Feb 2020

New exposé on the secretive AI group sheds light on Microsoft’s $1bn investment

Half of the $1bn Microsoft invested into OpenAI last year came in the form of Azure credits, according to a profile on the group published in MIT Technology Review.

In July, when Microsoft announced the mammoth investment into the AI research lab, which is dedicated to the safe and ethical creation of artificial general intelligence, Redmond said it would become the group’s “exclusive” provider of cloud computing services.

Microsoft’s investment came two months after OpenAI ditched its non-profit status by setting up a for-profit arm with a 100-fold limit on investor returns. The group said it needed more money to compete with others in the field.

However, a 6,000-word profile on the group written by data scientist and Tech Review reporter Karen Hao, has shed further light on the investment. According to Hao, the $1bn pot was “split between cash and credits to Azure”. At the time of writing, Microsoft has not responded to a request for comment.

“Compute-driven strategy”

Hao’s exposé, based on numerous interviews with past and present OpenAI employees, also reveals the degree to which the group’s AI research relies on a ‘muscle-mimic-car’ approach powered by the cloud, where AI capabilities are stretched by pumping greater computational resources into innovations developed in labs elsewhere.

Nowadays, companies obtain these computational resources by leasing high-powered servers and specialised GPUs in the cloud, which is why half of Microsoft’s investment was basically a ticket to soak up as much of the cloud’s power as possible.

According to Hao, who received unparalleled yet closely-guarded access to OpenAI’s inner sanctum, a group team called “Foresight” is dedicated to training existing algorithms with increasing amounts of data and computing power.

Hao claims that OpenAI has all-but decided that ploughing more compute and data into existing AI innovations, rather than working to develop a new AI paradigm, is the quickest path to achieving the holy grail of AGI.

Although co-founders Greg Brockman and Ilya Sutkever deny this is their sole strategy, Hao notes that most of OpenAI’s breakthroughs have arisen from this approach. She points to a recent paper posted by the Foresight team on open-source research database Arxiv, called “Scaling Laws for Neural Language Models”, which produced similar results to a paper posted months earlier.

For the leadership, the results of these experiments have confirmed its instincts that the lab’s all-in, compute-driven strategy is the best approach,” wrote Hao. “For roughly six months, these results were hidden from the public because OpenAI sees this knowledge as its primary competitive advantage. Employees and interns were explicitly instructed not to reveal them, and those who left signed nondisclosure agreements.”

OpenAI launched as a nonprofit research lab in 2015 and counted Elon Musk among its founders. Last February, Musk departed the group over “disagreements” about its development.

In November, OpenAI released a text-generation system that produced coherent text from simple prompts, which it trained on eight million text documents scraped from the web. The company didn’t initially release the system over concerns it could be misused.

Written by Tue 18 Feb 2020


agi Azure OpenAI
Send us a correction Send us a news tip